Inclusion criteria: age ≥ 65 years, and clinically diagnosed osteoporosis, i.e., a T-score ≤ −2.5 standard deviation at the lumbar spine and/or femoral neck (determined by DXA at max. 36 months prior to enrolment) and/or a history of osteoporotic fractures, i.e., a fracture of either radius, humerus, vertebra or femur secondary to a fall (low-energy trauma). Additional criteria were walking speed ≥ 0.8 m·s⁻¹, chair rising test ≤ 10 s, body mass index ≥ 20 kg/m², and at least one self-reported fall during the last 12 months. Exclusion criteria: active participation in a strength and/or endurance training during a 24-months period prior to enrolment, sports activity ≥ two times a week, cardiac insufficiency, cardiac pacemaker, chronic renal failure, non-curative treated tumors, systemic application of corticosteroids, peripheral arterial occlusive disease, poor mobility, i.e., walking more than 5 min not possible, severe osteoarthritis, rheumatism and/or severely impaired balance, planned absence of more than two weeks during the study period.

Clinical data on prescribed medications and history of disease were obtained by interview and chart review, respectively.

Group assignment (block randomization) was carried out using sealed, opaque envelopes with consecutive numbering and stratification according to sex and age. The person who opened the envelopes and carried out the group assignment was not involved in the generation and allocation concealment.

At baseline (t₀) as well as 3 and 6 months (t₁, t₂) after study onset, the performance capacity at the anaerobic threshold was determined by means of an incremental bicycle ergometer test [22]. Blood lactate was measured at each step, whereas heart rate was monitored continuously (Acentas, Germany) during the entire ergometry. From these data the anaerobic threshold was calculated. Continuous individual online heart rate monitoring (Acentas, Germany) was performed throughout the training sessions to keep the intensity at the anaerobic threshold during the loading phase and to ensure that the heart rate was below 90 bpm during the resting phase.

The intervention group trained twice a week for 6 months at a local sports club. Each session was supervised by more than one experienced instructor (KM, JIK, LS, AS, and RD). The exercises (modified step aerobics) were designed to increase strength, balance and agility. The patients were familiarized with a sequence of steps and movements (choreography) during the first 4 weeks. Each training session (60 min) started with a 10 min warm-up period consisting of basic aerobic steps without using the stepper, followed by step aerobics (30 min), balance training (10 min) and finally a 10 min cool-down period. The physical activities were accompanied by music. The patients danced a sequence of turns, forward, backward and side stepping with concomitant use of the arms and requirement of the stepper. For balance training the patients performed a sequence of parallel, tandem and one-legged stances with and without retroversion of the other leg and with eyes open and closed, respectively. During cool-down stretching and breathing exercises were performed. The control group was equipped with a step counter and was asked to maintain their individual level of physical activities.

Routine laboratory work-up prior at baseline was according to current guidelines and included alkaline phosphatase, parathyroid hormone, 25-hydroxyvitamin D₃, sex hormones, creatinine, calcium and phosphate [23]. Blood samples were analyzed with established laboratory procedures. Acceptance of the intervention and the osteoporosis-related quality of life was assessed by interview and questionnaires, respectively [24, 25]. Neuromuscular function was investigated at baseline and after 3 months and 6 months of training. All study examinations were done on days without exercise. Participants were asked to refrain from consuming alcohol and caffeine and not to perform any strenuous exercise the day before the measurement of neuromuscular function.

Transcutaneous electrical stimulation of the posterior tibial nerve in the popliteal fossa was used to generate the stimulus–response curves as described previously [27].

Briefly, the posterior tibial nerve was stimulated using a self-adhesive electrode (1 cm diameter). The anode was a self-adhesive electrode (40 × 90 mm, Spes Medica, Italy) fixed immediately distal to the patella on the anterior aspect of the knee. It was ensured that the electrical stimulation did not activate the tibialis anterior muscle. The percutaneous electrical stimuli were single pulses (1 ms duration, 400 V maximal voltage) delivered by a constant-current stimulator (Digitimer^® DS7A, Hertfordshire, UK). The inter stimulus intervals (ISI) were provided by a Digitimer^® train/delay generator (DG2A, Hertfordshire, UK). The testing procedure included random stimulation (ISI 7 s) with different current intensities, resulting in a recruitment curve. During this procedure, peak-to-peak maximal H-reflex (H_max) and maximal M-wave (M_max) of the soleus muscle were determined. M_max responses at rest and during MVC (M_sup) as well as V-wave responses during MVC were evoked with supramaximal stimulation intensity (140%).

Resting twitch torques were evoked using supramaximal single (1 ms duration, 400 V maximal voltage) and doublet stimuli (1 ms duration, 10 ms apart, 400 V maximal voltage). To normalize muscle activity of the tibialis anterior muscle, M_max amplitudes were evoked three times by stimulating the peroneal nerve close to the fibular head with supramaximal stimuli.

Surface EMG was recorded using bipolar EMG Ambu^® Blue Sensor N electrodes (2 cm diameter) as described previously [27]. The electrodes were firmly attached to the shaved, abraded and cleaned skin over soleus, medial gastrocnemius, lateral gastrocnemius and tibialis anterior of the right leg. The reference electrode was attached to the tibia of the ipsilateral leg. Signals were amplified (2500×), band-pass filtered (10-450 Hz) and digitized with a sampling frequency of 5 kHz through an analog-to-digital converter (DAQ Card™-6024E, National Instruments, USA). Both, the EMG and torque signals were sampled at 5 kHz and stored on a hard drive for later analysis with a custom built LABVIEW^® based program (Imago, Pfitec, Germany).

Torque signals as the result of muscle contraction were measured using a CYBEX NORM dynamometer (Computer Sports Medicine^®, Inc., Stoughton, MA). The individual positioning for each patient was similar throughout the examinations, i.e., they were seated with their knees straight and their feet firmly attached to the adapter of the dynamometer [27, 28] (Fig. 2). For this purpose, velcro straps and a snowboard binding were used and all participants wore shoes provided by the laboratory to guarantee the same material properties throughout all measurements. An ankle joint angle and hip joint angle of 90° and 80°, respectively (0° = full extension), were maintained during the sessions. The participants` ankle joints were aligned with the axis of the dynamometer. Straps across the thigh, waist and chest prevented excessive movements. Isometric maximum voluntary torque (iMVT) generation, i.e., the result of MVC, was tested by asking the participants to exert isometric maximal voluntary plantar flexions against the metal plate of the dynamometer for 3 s. For each trial, participants were thoroughly instructed to act as forcefully and as fast as possible. They were motivated by strong verbal encouragement and online visual feedback of the instantaneous dynamometer torque provided on a digital oscilloscope (HM1508, HAMEG Instruments, Germany). Care was taken that the MVC trials were performed without apparent countermovement or pre-tension (change of baseline torque <0.5 Nm during the 200 ms prior to contraction onset). A rest period of 1 min was allowed between the trials. Before MVC testing, the participants performed three to five MVC familiarization trials. When the coefficient of variance of three subsequent trials was below 5%, MVC testing was started. The participants performed seven to nine isometric maximal voluntary plantar flexions for V-wave testing. It has been shown that eight MVC attempts are sufficient to get reliable results for the evoked potentials [29].

The torque signals were corrected for the effect of gravity. The mean of the respective signals recorded during the MVC trials was used to calculate rate of torque development (RTD), iMVT and muscle activity. Explosive voluntary strength was determined by calculating the average RTD over time intervals of 0–50 and 0–200 ms relative to the onset of contraction.

Muscle activation during the early phase of contraction was analyzed by calculating the root mean square of the amplitude of the EMG signal (RMS-EMG) over time intervals of 0–50 and 0–200 ms relative to the onset of the EMG signals. RMS-EMG during iMVT (RMS-EMG_iMVT) was calculated over a 200 ms period at iMVT, i.e., 200 ms prior to the electrical stimulus. To reduce errors due to electrode relocation and between-session variability in skin impedance, subcutaneous fat and fascia muscle activity was normalized, i.e., RMS-EMG of soleus, medial gastrocnemius, lateral gastrocnemius and tibialis anterior were divided by their respective M_maxvalues (RMS-EMG/M_max). Furthermore, RMS-EMG/M_max was averaged across soleus, medial gastrocnemius, and lateral gastrocnemius to calculate triceps surae activation during the early phase of contraction and at iMVT (RMS-EMG_RTD/M_max and RMS-EMG_iMVT/M_max, respectively). Torque and EMG onsets were identified manually according to the method of Tillin et al. [30].

H_max, M_Hmax, M_max, M_sup and V-wave amplitudes were measured peak-to-peak and averaged, respectively. The H_max/M_max-ratio was used as a global index of modulations at the spinal level due to alterations in α-motoneuron excitability and/or presynaptic inhibition of primary muscle spindle afferents [31]. The M_Hmax/M_max-ratio was determined to ensure that the same proportion of α-motoneurons was activated by the electrical stimulation. The V/M_sup-ratio reflects the neural drive from spinal α-motoneurons to the soleus muscle [32]. Resting twitch torques were averaged and analyzed regarding their peak twitch torque, i.e., the highest value of twitch torque signal [27].

Recruitment rates, feasibility of inclusion-/exclusion criteria, the acceptance of the intervention (intensity per training session, frequency of training) and quality of life were evaluated ([24, 25]. The changes in neuromuscular function were used as preliminary estimates of effectiveness (proof-of-concept).

Data were checked for normal distribution using the Shapiro–Wilk test. The statistical approach comprised the analysis of covariance (ANCOVA) with baseline measurement and sex entered as covariates [33]. SPSS 20.0 (SPSS Inc., Chicago, IL, USA) was used for statistical analysis. Data obtained at baseline are presented as mean ± standard deviation, those obtained after 3 and 6 months of training are given as baseline-adjusted means ± baseline-adjusted standard deviation. If appropriate, data are presented as difference between means (95% confidence interval). All P values are two-sided and a P value below 0.05 was considered significant.

Patients were recruited during a three months period (late summer to autumn 2012), baseline examinations (4–6 participants per day) were performed during November 2012 and patients were randomized subsequently. The intervention was between December 2012 and June 2013. Individual study examinations were scheduled according to the protocol. Roughly, half of the participants from either group extended the training under the same supervisors for additional 6 months, although without concomitant examinations and/or follow-up examinations. The flow of participants through each stage of the study is given in Fig. 1.

Although virtually all patients were aware about the beneficial effects of physical activity, main arguments against participation were (1) care of spouse (n = 3), (2) distance/travel time to the training center (n = 3), (3) fear not to meet the expectations of the training group (n = 1), and (4) unwillingness to accept randomization (n = 3). Five patients from the control group refused to repeat neuromuscular testing after 3 and 6 months, mainly due to discomfort of the investigations and/or the expenditure of time. Drop outs during the study period were due to personal reasons and an unrelated injury. The training attendance rate as a rough measure for the acceptance of the intervention was 84%.

Obesity, hypertension, a history of curatively treated oncological disease, and minor cardiac, renal or lung problems were noticed in four, seven, five, five, three and one patient from the intervention group, respectively, as well as in five, ten, one, five, one and two patients from the control group. Two patients per group suffered from type 2 diabetes mellitus.